Skip to content

Releases: Saganaki22/Moss_TTS-ComfyUI

Release list

0.1.2

0.1.2 Latest
Latest

Choose a tag to compare

View all tags
@Saganaki22 Saganaki22 released this 28 Jun 15:51
v0.1.2
51e8a87

MOSS-TTS-ComfyUI v0.1.2

Fixed

  • Fixed MOSS-Audio-Tokenizer-v2 FlashAttention fallback under ComfyUI/AIMDO dynamic VRAM.
  • The codec now prefers the ComfyUI runtime device tag when entering inference autocast, instead of relying on the first parameter device.
  • This prevents the codec from incorrectly seeing dtype=torch.float32 and falling back to SDPA when FlashAttention is available.

Improved

  • Added a regression test to verify codec autocast uses CUDA BF16 when the model is CPU-staged but registered for CUDA runtime execution.

Validation

  • Tested with the ComfyUI virtual environment.
  • pytest tests -q
  • Result: 11 passed
Assets 2
Loading

0.1.1

0.1.1

Choose a tag to compare

View all tags
@Saganaki22 Saganaki22 released this 18 Jun 18:17
v0.1.1
3febfc7

Moss_TTS-ComfyUI v0.1.1

Initial public release for OpenMOSS MOSS-TTS Local Transformer v1.5 in ComfyUI.

Highlights

  • Added MOSS-TTS Local Transformer v1.5 support.
  • Added native 48 kHz stereo audio output.
  • Added direct text-to-speech generation.
  • Added zero-shot voice cloning from ComfyUI AUDIO.
  • Added speech continuation from prefix_audio plus prefix_text.
  • Added optional Whisper transcription node for continuation workflows.
  • Added support for MOSS language tags, pause markers, duration tokens, pronunciation markup, multilingual synthesis, and code-switching.
  • Added model catalog loader with download_if_missing.
  • Added dtype options: auto, bf16, fp16.
  • Added attention options: auto, sdpa, flash_attention, eager.

Model Management

  • Added native ComfyUI / AIMDO DynamicVRAM integration for:
    • MOSS-TTS main model
    • MOSS-Audio-Tokenizer-v2
    • optional Whisper ASR
  • Added real tensor registration for memory visualization.
  • Added unload and cleanup behavior when model, dtype, attention, or tokenizer files change.
  • Added AIMDO-safe native safetensors loading path.
  • Fixed dynamic tensor casting for embeddings, layer norms, conv layers, and codec codebooks.
  • Fixed Whisper AIMDO transcription issue caused by integer token IDs influencing embedding weight dtype.
  • Fixed punctuation-only Whisper hallucinations from being passed into continuation prompts.

Model Files

Large model weights are stored under:

ComfyUI/models/mosstts/

Bundled architecture, processor, tokenizer, and config assets are stored under:

ComfyUI/custom_nodes/Moss_TTS-ComfyUI/assets/

The main MOSS-TTS repository is not fully snapshot-downloaded into models/. Only required model weight files are downloaded when needed.

Included Assets

  • Bundled MOSS-TTS Local Transformer v1.5 architecture/config/tokenizer assets.
  • Bundled MOSS-Audio-Tokenizer-v2 architecture/config assets.
  • Included example_workflows/ with workflow JSON and preview image.
  • Included .github/ workflow folder.
  • Included English and Simplified Chinese READMEs.

Verification

Tested with the ComfyUI venv:

  • Python compile check passed.
  • Runtime management tests passed: 10 passed.
  • AIMDO MOSS main model smoke test passed.
  • AIMDO codec encode/decode smoke test passed.
  • AIMDO Whisper speech transcription smoke test passed.

Links

Loading